Setting and Subjects
This study was conducted in a 51-bed palliative care unit in Japan from April 2018 to October 2019. Palliative care physicians enrolled patients with an estimated life expectancy of a few weeks, as determined by clinical judgment and prognostic models, such as the Palliative Prognostic Index23 and Palliative Prognosis Score.24 Patients were included in this study if their estimated survival period till death was 3 days or more.
Dependent variable (outcome)
The outcome variable was the occurrence of death every 24 hours. Death occurrence was defined as the patient’s medical record stating a death diagnosis by a doctor in charge. Moreover, one day prior to death was defined as the time of death to 24 h before death.
Main independent variables measured by a monitor
This study used a non-wearable actigraphy monitor (Nemuri SCAN, Paramount Bed Co. Ltd., Tokyo, Japan) containing a highly sensitive pressure sensor.25 The device was 28.6 cm long, 77 cm wide, and 2.5 cm thick (Figure 1). The sensing monitor was placed under a mattress. This monitor measured respiratory rate, heart rate, and the amount of activity per minute and calculated sleep, arousal, and out-of-bed status using these data. The reliability and validity of the Nemuri SCAN are guaranteed (Figure1).21,22
In this study, we focused on two vital signs, respiratory rate and heart rate, because these data were automatically gathered every minute by the Nemuri SCAN device, and created a dataset of the mean, maximum, minimum, and variance of respiratory rate and heart rate every 24 hours up to death for each patient, each calculated from monitoring data every minute.
We examined the mean, maximum, minimum, and variance values of respiratory rate and heart rate measured by monitoring on the day of death (n=24) and other days (2–14 days before death, n=216).
Other independent variables extracted by medical records
Age, sex, and primary cancer site were recorded as patient demographic characteristics.
We also collected the following independent variables that were evaluated every 24 hours up to death, which were extracted from the patient’s medical records, referring to a previous study23: (1) nurse-measured vital signs were collected, including respiratory rate, heart rate, temperature, systolic blood pressure, and diastolic blood pressure, and these vital signs were used as continuous values; moreover, nurses measured vital signs twice a day, and we recorded the mean, maximum, and minimum for that day; (2) subjective symptoms comprising six items: pain, dyspnea, malaise, restlessness, nausea, and dry mouth; (3) objective symptoms comprising eight items: painful expression, cough, sputum, cold limbs, edema, oral fluid intake, oral food intake, and delirium; and (4) drug use that was divided into opioids, non-opioids, hypnotics, and artificial hydration.
Two nurses (IK and SF) read each medical record and extracted clinical data separately. Thereafter, the two confirmed the validity of the data obtained.
We created a dataset of (i) outcomes (from death-day to 2–14 days before death per patient, totaling 240 patient-day data), (ii) data measured by a monitor (mean, variance, maximum, and minimum respiratory rate and heart rate), and (iii) patient demographic characteristics, symptoms, and drug use from medical records. Symptoms and drug use were integrated in chronological order every 24 h to death.
First, we calculated the descriptive statistics for the patient demographic characteristics, nurse-measured vital signs, subjective symptoms, objective symptoms, and drug use to clarify the background and clinical status of the patients.
Next, by univariate analyses using t-tests, we examined the association between the value of death day (n=24) and the value of other days (2–14 days before death, n=216) for the mean, maximum, minimum, and variance of respiratory rate and heart rate measured by the monitor.
Finally, we examined the association between respiratory and heart rates and death using repeated measures logistic regression analysis. To this end, we prepared three types of binary outcomes for each day: death occurrence within 24 hours, 48 hours, and 72 hours. Several explanatory variables representing the within-day level/variation of the respiratory and heart rates were prepared, including the mean, maximum, minimum, and variance of respiratory rate and heart rate. Note that each outcome was defined every 24 hours from the date of death, and each patient had multiple observations. Accounting for within-patient correlation among repeatedly observed outcomes, we performed a repeated measures logistic regression analysis with the generalized estimating equation (GEE).26 In making a valid statistical inference, we utilized the robust sandwich variance with the working independent correlation. According to the fitted logistic model, we calculated the predicted probability of death within the next 3 days (72 hours) based on the respiratory and heart rates for each of the 24 subjects.
All analyses were performed using the Statistical Analysis System (SAS) version 9.4 (SAS, Institute Inc., Cary, NC, USA), and p<0.05 was considered significant.
Research ethics and patient consent
This study was conducted with the approval of the Ethics Review Committee of Intervention Studies and Observational Research, Osaka University Hospital (approval number: 1741110). After explaining the study protocol to eligible patients, written informed consent was obtained from each patient. Participation in the study was voluntary, and patients were informed that all data would be anonymous and their privacy and personal information would be protected. If a patient did not have sufficient mental capacity to decide on study participation, written consent was obtained from the patient’s family/proxy. All methods were performed in accordance with Di-CHiLD project guideline, which is the Osaka University and Daikin Industries collaborative study.